Introduction: The Secret Behind Small Teams and Big Results

Vikas Paruchuri introduces himself as CEO of Datalab and discusses how a team of just 3 people achieved 40,000 GitHub stars, 7-figure ARR (Annual Recurring Revenue), and built state-of-the-art AI models. He explains that over the past year, he personally trained multiple models, founded the company, raised seed funding, and only recently hired the first employee, bringing the team to 4 people.

"I trained these models over the past year, founded the company, and raised seed funding. I didn't sleep much. That turned out to be important."

Since hiring its first employee in January 2023, Datalab has grown revenue 5x, with customers including top AI research labs, universities, Fortune 500 companies, and AI startups. The core of this talk is sharing how they grew so fast with a small team and their team-building philosophy.

The Misconception About Headcount and Productivity

Vikas challenges the common Silicon Valley belief that "hiring more people means more productivity."

"In Silicon Valley, there's a fixed idea that if you raise money and hire lots of people, you can build more. But in my experience, that almost never works out."

He shares his experience at his previous company, DataQuest, where he grew the team to 30, then went through two rounds of layoffs down to 7. Surprisingly, productivity and happiness actually increased after the layoffs.

"A few months after cutting the team, we were actually far more productive. That made me wonder why."

He offers four hypotheses for this phenomenon:

• Over-hiring specialists: As the team grows, roles become too fragmented, preventing flexible problem-solving
• Remote work limitations: Requires intentional processes and synchronization that consume significant time
• Meeting overload: Middle managers multiply, meetings increase, and actual work time shrinks
• Senior talent inefficiency: Senior people spend their time managing juniors rather than exercising their core skills

The Golden Age and Organizational Growth Curves

Vikas describes the "golden age" that most companies experience -- the period when everyone is aligned and building the core product. After that, the organization grows, departments and roles multiply, and bureaucracy and priority confusion set in.

"Google building search, Microsoft building Windows -- that was the golden age. After that, the organization grows and people get trapped in small boxes."

He asks "Why does this golden age have to end?" and shares insights gained from working with Jeremy Howard of Answer AI:

• Build a team of fewer than 15 generalists (multi-talented people)
• Fill gaps with AI and internal tools
• Use simple, proven technology -- nothing complex

"Build a team of fewer than 15 generalists, and fill the rest with AI and internal tools. That's the core."

This approach requires a high-trust, customer-centric culture where every team member communicates directly with customers and participates in building the product.

Case Study: State-of-the-Art Models Built by a Tiny Team

Vikas describes the development of the recently released Syria OCR 3 model. This model boasts 500 million parameters, support for 90 languages, and 99% accuracy, including capabilities not found in existing models.

"This model supports 90 languages and scored 99% accuracy on internal benchmarks. It even handles math problems."

Just two people handled the entire process -- from customer requirements discovery -> paper research -> architecture design -> data pipeline construction -> model training -> inference code -> product integration.

• What would require multiple teams at a large company was handled end-to-end by a tiny crew
• Minimized information loss and inefficiency from handoffs between teams
• Enabled fast feedback loops and tight integration

"What would typically be split across multiple teams at a normal company, we did end-to-end with two people. That's why it was much faster and more efficient."

Operating Principles for an Elite Small Team

Vikas reiterates that "more people doesn't mean more productivity" and explains concretely how they operate.

1. Hire Senior Generalists

• What matters isn't years of experience but maturity and ownership in problem-solving
• People who communicate directly with customers and persist until the problem is solved

"Senior doesn't mean tenure. It means seeing a problem and saying 'I'll solve this' -- that attitude and maturity."

2. Eliminate Unnecessary Complexity

• Don't obsess over cutting-edge technology; choose the simplest approach
• Example: A single machine with shell scripts instead of a complex Kubernetes cluster

"Instead of a complex Kubernetes cluster, just running on one machine with shell scripts is often better."

3. Prefer In-Person Collaboration

• Prefer in-office work for fast feedback and collaboration
• Remote work can slow things down due to increased process overhead

4. Simplify Architecture and Maximize Reuse

• Maximize component reuse for easy maintenance and scaling
• Use lightweight frontend tools like server-rendered HTML, HTMX, and Alpine
• Design modular code so AI can easily contribute

"We don't use complex frameworks like React. We stick to server-rendered HTML and lightweight libraries."

5. Minimize Process, Build Trust-Based Culture

• Eliminate unnecessary management and bureaucracy
• Only hire people who can drive themselves

"If someone seems like they'll need a lot of managing, we just don't hire them."

Filling Organizational Gaps with AI

As a document processing AI company, Datalab serves customers with varying document parsing requirements. Previously they'd send field engineers for custom development, but now AI models can replace this role.

• Models iteratively learn and optimize for diverse customer requirements
• AI resolves complexity instead of on-site personnel

"AI models can now iteratively learn and optimize for diverse customer needs. It's solving complexity with AI instead of field staff."

When this model will hit its limits remains unknown, but the key is choosing wisely as you grow.

"I don't know if this model will work forever, but it's ultimately a choice. You can hire more people, or you can choose a more efficient approach."

Role Structure and Culture

Datalab's roles are divided into three: research engineer, full-stack engineer, and go-to-market (sales/marketing/support combined), but everyone communicates with customers and participates in product development.

• No politics or selfishness -- focus only on work, colleagues, and customers
• Market-leading compensation
• Prefer self-motivated talent only
• Value low ego and execution (GSD -- Get Shit Done)

"Politics is death for small teams. We want people who think only about work, colleagues, and customers."

When hiring, he emphasizes the patience to wait for the best rather than rushing to fill a seat.

"My worst hires came when I rushed. My best hires came when I waited for 'this is the one.'"

Scaling Productivity: Efficiency Over Headcount

As the company grows, he proposes scaling productivity through efficiency rather than headcount:

1. Raise salary bands: Hire more experienced talent for the same roles
2. Invest in compute resources: Provide more GPUs and other resources
3. Leverage AI tools: Aggressively adopt tools that maximize productivity

"One researcher with 64 GPUs is far more productive than one with 8."

Q&A: Real Experience and Hiring Process

In the Q&A, he explains that after layoffs, they maintained the product's essence while aggressively cutting less important features.

"When you have too many people, there's not enough work, and you end up building pointless features. With a small team, you focus on what really matters."

On transforming large-company culture: "Those organizations are hard to change. You're better off starting a small company and doing it better."

"Organizations with entrenched culture are hard to change. You're better off starting a small company that does it better."

Hiring is done through open source and Twitter activity, with the hiring process as follows:

1. Casual conversation: Discuss problems together like actual colleagues
2. Paid project: 10 hours, $1,000 compensation, real collaboration experience
3. Culture fit interview: Verify team chemistry

"The paid project is the best way to actually work together. $1,000 isn't much, but it helps tremendously in finding the right person."

Ultimately, they hired about 40% of candidates who reached the interview stage.

Conclusion: The Future of Small Teams with Big Impact

Vikas reiterates the message that "more people doesn't automatically mean more productivity" and invites anyone interested in building a culture where small teams move fast and efficiently to reach out.

"If this kind of culture interests you, reach out anytime. I'd love to talk."

Key Concepts:

• Elite Small Team
• Generalist
• AI and Internal Tool Utilization
• Simple Technology, Modularization
• Customer-Centric, Trust-Based Culture
• Scaling Productivity (Efficiency Over Headcount)
• Execution (GSD), Low Ego
• Paid Project-Based Hiring

Datalab demonstrates that even as a small team, by maximizing efficiency, execution, and the power of AI, they can create massive impact.